The long-term goal of this proposal is to predict protein structures and protein-protein complex structures to facilitate the development of therapeutic drugs. This proposal addresses the urgent need for a more accurate energy function for high-resolution protein-structure prediction and protein-protein interaction prediction. Currently, there are three complementary approaches to this problem, based on: physical principles (physical-based), known protein structures (knowledge-based or statistical), or empirical methods. Among the three, establishing a statistical energy function at an "all-atom" level of detail is the least explored approach. Here we propose a statistical energy function built on a mixture of atoms and molecular fragments, rather than on atoms alone. Inclusion of molecular fragments accounts for many interactions missed partially or wholly by commonly used atom-based approaches. Preliminary studies have shown a multi-fold improvement in the accuracy and specificity of refolding completely unfolded segments with secondary structure elements. This success is a preview of the potential of the proposed fragment-based approach to statistical energy functions. PUBLIC HEALTH RELEVANCE Many diseases including Alzheimer's disease, cystic fibrosis, and Bovine Spongiform Encephalopathy (Mad Cow disease) are caused by malfunction of the nanomachines called proteins. The energy function that governs the function of these proteins has yet to be discovered. Here, we propose to uncover the energy function by developing a fragment-based statistical approach. Successful completion of this project should allow us to more accurately predict protein structures based on their gene-specified sequence information. Knowing a protein's structure is essential for understanding its function and developing therapeutic drugs to block or activate the protein's function.